Research Article Time-Aware IoE Service Recommendation on …downloads.hindawi.com/journals/misy/2016/4397061.pdf · 2019-07-30 · 2016.01.14 2016.06.09 User target 1 1 1 1 2 5 4

Research ArticleTime-Aware IoE Service Recommendation on Sparse Data

Lianyong Qi12 Xiaolong Xu3 Wanchun Dou2 Jiguo Yu1 Zhili Zhou3 and Xuyun Zhang4

1School of Information Science and Engineering Qufu Normal University Qufu China2Department of Computer Science and Technology State Key Laboratory for Novel Software Technology Nanjing UniversityNanjing China3School of Computer and Software Nanjing University of Information Science and Technology Nanjing China4Department of Electrical and Computer Engineering University of Auckland Auckland New Zealand

Correspondence should be addressed to Lianyong Qi lianyongqigmailcom

Received 15 September 2016 Accepted 10 November 2016

Academic Editor Beniamino Di Martino

Copyright copy 2016 Lianyong Qi et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

With the advent of ldquoInternet of Everythingrdquo (IoE) age an excessive number of IoE services are emerging on the web whichplaces a heavy burden on the service selection decision of target users In this situation various recommendation techniquesare introduced to alleviate the burden for example Collaborative Filtering- (CF-) based recommendation Generally CF-basedrecommendation approaches utilize similar friends or similar services to achieve the recommendation goal However due to thesparsity of user feedback it is possible that a target user has no similar friends and similar services in this situation traditional CF-based approaches fail to produce a satisfying recommendation result Besides recommendation accuracywould be decreased if timefactor is overlooked as IoE service quality often varies with time In view of these challenges a time-aware service recommendationapproach named Ser Rectime is proposed in this paper Concretely we first calculate the time-aware user similarity afterwardsindirect friends of the target user are inferred by Social BalanceTheory (eg ldquoenemyrsquos enemy is a friendrdquo rule) finally the servicespreferred by indirect friends of the target user are recommended to the target user At last through a set of experiments deployedon dataset WS-DREAM we validate the feasibility of our proposal

1 Introduction

With the wide adoption of various smart devices and connec-tion technologies human society is gradually transforminginto an ldquoInternet of Everythingrdquo (IoE) one [1ndash5] In the ageof IoE the links among users devices or other things couldbe built easily which significantly improves peoplersquos qualityof life and also brings many challenging open problems thatneed to be addressed [6ndash11]

With the advent of IoE age an excessive number ofIoE services with different functionalities or qualities areemerging on the web which places a heavy burden on theservice selection decision of target users [12ndash14] In this situ-ation various recommendation techniques are put forward tohelp alleviate the service selection burden for example Col-laborative Filtering- (CF-) based recommendation [15ndash19]Generally CF-based recommendation approaches (includinguser-based CF item-based CF and hybrid CF) utilize thesimilar friends or similar services of target users to achieve

the recommendation goal However in certain situations theavailable user-service rating data generated from historicalservice invocations is really sparse [20] Therefore it isprobable that a target user cannot find hisher similar friendsand similar services of target services (here target servicesmean the services preferred by target users) In this situationtraditional CF-based recommendation approaches cannotreturn a satisfying recommendation result which brings agreat challenge for the robustness of service recommendationapproaches Besides the quality of an IoE service often varieswith time due to the unstable network environment [21]For example the response time of a ticket-order serviceoften becomes larger when Christmas day is approachingTherefore recommendation accuracy would be decreased ifthe time factor is not taken into consideration

In view of the above two challenges a novel time-awareservice recommendation approach that is Ser Rectime isput forward in this paper to make robust and accurateservice recommendation for target users when the historical

Hindawi Publishing CorporationMobile Information SystemsVolume 2016 Article ID 4397061 12 pageshttpdxdoiorg10115520164397061

2 Mobile Information Systems

user-service rating data is sparse Concretely in Ser Rectimewe first calculate the time-aware user similarity afterwardswe look for the enemies (antonym of ldquofriendrdquo) of the targetuser and further determine the ldquoindirect friendsrdquo of targetuser by Social Balance Theory [22] (eg ldquoenemyrsquos enemy is afriendrdquo rule) finally the services preferred by indirect friendsof target user are recommended to the target user

The contributions of our paper are threefold

(1) Time factor is considered in user similarity calcula-tion to adapt to the dynamic quality variation of IoEservices which makes the subsequent recommenda-tion result more objective and accurate

(2) Social Balance Theory is introduced for service rec-ommendation on sparse data so that the recommen-dation robustness is improved

(3) A wide range of experiments are designed anddeployed on a real web service quality set WS-DREAM [23] so as to further validate the feasibilityof our proposal

The remainder of our paper is organized as follows InSection 2 we first formalize the CF-based service recommen-dation problem and afterwards demonstrate the motivationof our paper In Section 3 a novel time-aware servicerecommendation approach named Ser Rectime is broughtforth A set of experiments are deployed in Section 4 andevaluations are presented in Section 5 Finally in Section 6we summarize the paper and point out our future researchdirections

2 Formalization and Motivation

In this section we first formalize theCF-based service recom-mendation problemAnd afterwards an example is presentedto demonstrate the motivation of our paper intuitively

21 Formal Specification Generally the CF-based servicerecommendation problem could be specified with a four-tuple CF-Ser-Rec (119880WSrarr usertarget) where

(1) 119880 = user1 user119898 denotes the user set in user-service invocation network and 119898 is the number ofusers

(2) WS = ws1 ws119899 denotes the web service set inuser-service invocation network and 119899 is the numberof web services

(3) rarr= (user119894 119877119894minus119895997888997888997888rarr119879119894minus119895

ws119895) | 1 le 119894 le 119898 1 le 119895 le 119899 de-notes the historical invocation record set that isuser119894

119877119894minus119895997888997888997888rarr119879119894minus119895

ws119895 whichmeans that user119894 invoked ws119895 in

the past and the rating of ws119895 by user119894 is119877119894minus119895 Here forsimplicity the well-known 1lowast 2lowast 3lowast 4lowast 5lowast ratingsystem is adopted to depict119877119894minus119895 Also each invocationrecord owns a timestamp 119879119894minus119895 that indicates the ser-vice invocation time (here we assume the timestampformat is ldquoyyyymmddrdquo)

(4) usertarget denotes the target user that requires servicerecommendation and usertarget isin 119880 holds

With the above formalization CF-based service recom-mendation problem could be specified as follows accordingto the historical invocation record set ldquorarrrdquo between users (in119880) and web services (in WS) find out the services that werenever invoked but may be preferred by target user usertargetand recommend them to usertarget

22 Motivation In this subsection the paper motivation isclarified more intuitively with the example shown in Figure 1In Figure 1 there are three users TomAliceBob (Tom isthe target user) in set 119880 and six web services ws1ws2ws3ws4ws5ws6 in set WS historical user-service invocationrecord set rarr (including rating data and timestamp) is alsopresented in Figure 1

Then according to the Adjusted Cosine Similarity [24](ie ACS isin [minus1 1] here the reason that we utilize ACSfor similarity calculation is that user-service rating data isoften discrete and the rating scales of different users oftenvary) we can calculate the similarity between target userTom and other two users (ie Alice and Bob) ConcretelySim(TomAlice) = minus02747 and Sim(TomBob) = Null (asno services were rated by both Tom and Bob) Therefore wecan conclude that target user Tom has no similar friends soaccording to the traditional user-based CF recommendationapproaches no qualified services are recommended to Tom

Likewise the similarities between target services (ie ws1andws2) and other services (ie ws3 ws4 ws5 andws6) couldalso be obtained Concretely Sim(ws1ws3) = Sim(ws1ws4)= Sim(ws2ws3) = Sim(ws2ws4) = minus1 and Sim(ws1ws5) =Sim(ws1ws6) = Sim(ws2ws5) = Sim(ws2ws6) = Null (asno users invoked any pair of services above simultaneously)Therefore a conclusion could be drawn that target user Tomrsquospreferred services (ie ws1 and ws2) have no similar servicesso according to the traditional item-based CF recommenda-tion approaches no qualified services are recommended toTom

Therefore in this situation traditional CF-based ser-vice recommendation approaches (eg user-based CF item-based CF or hybrid CF) cannot make accurate service rec-ommendation for target user Besides as Figure 1 shows thetimestamps of various invocation records are often differentso the service recommendation accuracy and fairness wouldbe decreased if we overlook the time factor in recommen-dation process In view of the above two challenges a novelservice recommendation approach named Ser Rectime is putforward in Section 3 Ser Rectime cannot only find outthe indirect friends of a target user so as to improve therecommendation robustness in sparse-data environment butalso consider time factor in recommendation so as to ensurethe fairness and accuracy of recommendation results

3 Time-Aware Service RecommendationApproach Ser_Rectime

In this section a novel time-aware recommendation ap-proach that is Ser Rectime is introduced to deal with the

Mobile Information Systems 3

Tom Alice Bob

User level

Service level

20100101 20120407 20110217

20140927 20130722

20151126 20150116

20130616 2016011420160609

Usertarget

1lowast

1lowast

1lowast1lowast

2lowast

4lowast5lowast

5lowast

5lowast

5lowast

ws1 ws2 ws3 ws4 ws5 ws6

Figure 1 Time-aware service recommendation when target user has no similar friends and similar services

service recommendation problem on sparse data The mainidea of our proposal is as follows first we calculate thetime-aware similarity between target user and other usersbased on the historical user-service rating records secondaccording to the derived user similarity and Social BalanceTheory we infer the indirect friends of target user thirdthe services preferred by indirect friends of target user arerecommended to the target user Concretely the three steps oftime-aware service recommendation approach (Ser Rectime)are as follows

Step 1 (time-aware user similarity calculation) Calculate thesimilarity Sim(usertarget user119894) between usertarget and otherusers user119894 isin 119880 If 119878im(usertarget user119894) le 119875 (119875 is similaritythreshold) then user119894 is considered as an enemy of usertarget

Step 2 (determining indirect friends of target user) Accord-ing to the enemies (derived in Step 1) of target user and SocialBalance Theory determine target userrsquos indirect friends

Step 3 (service recommendation) Select the services pre-ferred by the indirect friends (derived in Step 2) of target userand recommend them to the target user

The three steps are explained as follows

Step 1 (time-aware user similarity calculation) In this stepwe calculate the similarity between target user usertarget andother users user119894 (user119894 isin 119880) that is Sim(usertarget user119894) Asuser-service rating data is often discrete and rating scores ofa service by different users are often varied Adjusted CosineSimilarity is recruited here for user similarity calculationConcretely Sim(usertarget user119894) could be obtained based on

Sim (usertarget user119894)

= sumws119895isin119868 (119877targetminus119895 minus 119877target) lowast (119877119894minus119895 minus 119877119894)radicsumws119895isin119868target (119877targetminus119895 minus 119877target)2 lowast radicsumws119895isin119868119894 (119877119894minus119895 minus 119877119894)2

(1)

Here set 119868 denotes the common services that were ratedby both usertarget and user119894 sets 119868target and 119868119894 denote theservice set rated by usertarget and user119894 respectively 119877targetminus119895and 119877119894minus119895 denote ratings of service ws119895 by usertarget and user119894respectively while 119877target and 119877119894 represent usertargetrsquos and

user119894rsquos average rating values Then according to (1) we cancalculate the similarity between target user and any otheruser

Next we improve the user similarity formula in (1) byintroducing four kinds of weight coefficients

(i) Weight for Service Intersection Size As formula (1)indicates 119868target and 119868119894 denote the service set ratedby usertarget and user119894 respectively Generally forusertarget and user119894 the larger their service intersec-tion (ie set 119868 in formula (1)) is the more convincingtheir similarity is So in order to depict this correla-tion weight119882ser-intersection (in formula (2)) is assignedto user similarity Sim(usertarget user119894)119882ser-intersection = 12 lowast (

119868target cap 119868119894119868target + 119868target cap 119868119894119868119894 ) (2)

(ii) Weight for Invocation Time Due to the dynamicand unstable network environment the IoE servicequality is often varied with time therefore twoneighboring service invocationswith close invocationtime often contribute more to the user similarity Inview of this observation similar to work [4] weight119882time in formula (3) is assigned to user similaritySim(usertarget user119894) In (3) 120572 (120572 ge 0) is a parameterwhile 119879targetminus119895 and 119879119894minus119895 represent the invocation timeof service ws119895 by usertarget and user119894 respectively

119882time = 119890minus120572|119879targetminus119895minus119879119894minus119895| (3)

(iii) Weight for Invocation Load Service invocation timecannot reflect all the time-related information insimilarity calculation For example user1 and user2invoked the same ticket-booking service on 22-12-2015 and 24-12-2015 respectively Although theirservice invocation time is close the user experi-enced service quality may vary significantly as heavyservice load is inevitable when Christmas day isapproaching So in order to depict the effect of serviceload on user similarity weight 119882load in formula (4)is assigned to user similarity Sim(usertarget user119894)In (4) Loadtargetminus119895 and Load119894minus119895 denote the service


loads when usertarget and user119894 invoked service ws119895respectively

119882load = 1 minus10038161003816100381610038161003816Loadtargetminus119895 minus Load119894minus11989510038161003816100381610038161003816

max Loadtargetminus119895 Load119894minus119895 (4)

(iv) Weight for Service Version In the life cycle of aweb service service provider may publish a seriesof service versions with updated service qualitiesTherefore if two users invoked an identical webservice that belongs to different versions it wouldnot make much sense to compare their experienced

service qualities In view of this observation weight119882version is suggested as follows

119882version

= 1 usertarget user119894 invoked same service of same version

0 else(5)

With the above analyses we can update the user similaritySim(usertarget user119894) in (1) to be time-aware user similaritySimtime(usertarget user119894) in (6) Then based on (6) we cancalculate the time-aware similarity between target user andany other user

Simtime (usertarget user119894) = 119882ser-intersection lowast sumws119895isin119868119882time lowast119882load lowast119882version lowast (119877targetminus119895 minus 119877target) lowast (119877119894minus119895 minus 119877119894)radicsumws119895isin119868target (119877targetminus119895 minus 119877target)2 lowast radicsumws119895isin119868119894 (119877119894minus119895 minus 119877119894)2

(6)

Furthermore according to the derived user similaritythe enemy set of target user that is Enemy set (usertarget)could be obtained based on (7) Here parameter 119875 (minus1 le119875 le minus05) is a predefined user similarity threshold for enemyrelationship

user119894isin Enemy set (usertarget) if Simtime (usertarget user119894) le 119875notin Enemy set (usertarget) else (7)

Step 2 (determining indirect friends of target user) InStep 1 we have obtained the enemy set of target userthat is Enemy set (usertarget) Next in this step we willintroduce how to get the indirect friends of target userthat is Indirect friend set(usertarget) based on the obtainedEnemy set (usertarget) and Social BalanceTheory First of allwe introduce Social Balance Theory briefly

Social Balance Theory [22] was first put forward bypsychologist F Heider in 1958 The theory investigates thestable social relationships among involved three parties (ie119875 119874 and 119883 in Figure 2) Concretely according to Figure 2we introduce the four stable social relationships respectivelyin a more intuitive manner

(a) Friendrsquos Friend Is a Friend If119883 is a friend of 119874 and 119874is a friend of 119875 then we can infer that 119883 is probablyan indirect friend of 119875 (see Figure 2(a))

(b) Enemyrsquos Enemy Is a Friend If119883 is an enemyof119874 and119874is an enemy of 119875 then we can infer that119883 is probablyan indirect friend of 119875 (see Figure 2(b))

(c) Friendrsquos Enemy Is an Enemy If 119883 is an enemy of 119874while 119874 is a friend of 119875 then we can infer that 119883 isprobably an indirect enemy of 119875 (see Figure 2(c))

(d) Enemyrsquos Friend Is an Enemy If 119883 is a friend of 119874while 119874 is an enemy of 119875 then we can infer that 119883is probably an indirect enemy of 119875 (see Figure 2(d))

Next with the above four rules in Social Balance Theorywe introduce how to obtain the indirect friend set of target

user that is Indirect friend set(usertarget) based on thederived Enemy set(usertarget) in Step 1

Concretely for each user119894 isin Enemy set (usertarget) wefirst calculate hisher similarities with other users based on(6) afterwards we determine user119894rsquos enemies user119896 (ieuser119896 isin Enemy set (user119894)) based on (7) and user119894rsquos friendsuser119892 (ie user119892 isin Friend set (user119894)) based on (8) In(8) parameter minus119875 denotes the user similarity threshold forfriend relationship To ease the understanding of readersthe relationships among usertarget user119894 user119896 and user119892 arepresented in Figure 3

user119892isin Friend set (user119894) if Simtime (user119894 user119892) ge minus119875notin Friend set (user119894) else (8)

Then according to ldquoenemyrsquos enemy is a friendrdquo rule in Fig-ure 2(b) we can infer that user119896 is probably an indirect friendof usertarget and the probability could be calculated basedon (9) If Probabilityfriend(usertarget user119896) ge minus119875 then user119896is regarded as a qualified indirect friend of target user andput into set Indirect friend set(usertarget) (see Figure 3(b))Likewise according to ldquoenemyrsquos friend is an enemyrdquo rulein Figure 2(d) it can be inferred that user119892 is probablyan indirect enemy of usertarget and the probability can beobtained based on (10) If Probabilityenemy(usertarget user119892) geminus119875 then user119892 is regarded as a qualified indirect enemyof target user and put into set Enemy set (usertarget) (seeFigure 3(d))

Probabilityfriend (usertarget user119896)= Simtime (usertarget user119894)lowast Simtime (user119894 user119896)

(9)


P O

Friend relationshipEnemy relationship

X

(a)

P O


X

(b)

X

P O


(c)

X

P O


(d)

Figure 2 Four stable relationships among 119875119883 and119874 according toSocial BalanceTheory (the blue arrows between 119875 and119883 denote theinferred relationships)

Probabilityenemy (usertarget user119892)= 10038161003816100381610038161003816Simtime (usertarget user119894)lowast Simtime (user119894 user119892)10038161003816100381610038161003816

(10)

Next for each user119909 isin Indirect friend set(usertarget) wecalculate hisher similarity with other users (isin (119880-usertarget-Indirect friend set(usertarget)-Enemy set(usertarget))) basedon (6) and further determine user119909rsquos enemies user119910 based on(7) and user119909rsquos friends user119911 based on (8) Then accordingto ldquofriendrsquos enemy is an enemyrdquo rule in Figure 2(c)user119910 is regarded as an indirect enemy of target user andput into Enemy set (usertarget) (see Figure 3(c)) if theprobability in (11) is larger than minus119875 Likewise accordingto ldquofriendrsquos friend is a friendrdquo rule in Figure 2(a) user119911is considered as an indirect friend of target user and putinto Indirect friend set (usertarget) (see Figure 3(a)) if theprobability in (12) is larger than minus119875

Probabilityenemy (usertarget user119910)= 10038161003816100381610038161003816Probabilityfriend (usertarget user119909)lowast Simtime (user119909 user119910)10038161003816100381610038161003816

(11)

Probabilityfriend (usertarget user119911)= Probabilityfriend (usertarget user119909)lowast Simtime (user119909 user119911)

(12)

Repeat (a)ndash(d) process in Figure 3 until setIndirect friend set(usertarget) stays stable Then we

obtain the indirect friends of target user that isIndirect friend set(usertarget)Step 3 (service recommendation) In Step 2 wehave obtained target userrsquos indirect friend setIndirect friend set(usertarget) Next in this step weselect the services preferred (ie with 4lowast or 5lowast rating)by indirect friends of target user and recommend themto the target user More formally for each elementuser119909 isin Indirect friend set(usertarget) if hisher rating overweb service ws119895 (1 le 119895 le 119899) that is 119877119909minus119895 isin 4lowast 5lowast holdsthen ws119895 is put into the recommended service set that isRec Ser Set Finally all the web services in set Rec Ser Setare recommended to usertarget

With above Step 1ndashStep 3 of our proposed Ser Rectimeapproach a set of IoE services (in set Rec Ser Set) are recom-mended to the target user More formally the pseudocode ofSer Rectime(119880WSrarr usertarget) is presented in Algorithm 1(please note that algorithm Ser Rectime(119880WSrarr usertarget)is abbreviated as Ser Rectime in the whole paper)

4 Experiment

In this section a set of experiments are designed and tested tovalidate the feasibility of our proposed Ser Rectime approachin terms of recommendation accuracy recall and efficiency

41 Experiment Dataset and Deployment The experiment isbased on a real service quality dataset WS-DREAM [23]WS-DREAM consists of 4532 IoE services on the web and142 distributed users from Planet-Lab are employed forevaluating the real quality (eg response time and throughput)of services in 64 time intervals (time interval = 15 minutes)

Our paper focuses on the user-service rating-based rec-ommendation while available user-service rating data isreally rare on the web therefore in the experiment we needto transform the service quality data in WS-DREAM intocorresponding user-service rating data (essentially objectiveservice quality data and subjective user-service rating databoth reflect the service running quality therefore we arguethat the transformation from former data to latter datamakes sense for the service recommendation simulationhere) Concretely the transformation process is as follows wedetermine the minimal and maximal quality values (denotedby min and max resp) of a service observed by an identicaluser and afterwards we divide the range [minmax] intofive subranges in an arithmetic progression manner eachcorresponding to a rating value For example if the minimaland maximal throughput values of a service ws119895 observed byuser119894 are 10 kbps and 60 kbps respectively thenwe can get fivesubranges after division that is [10 20) kbps [20 30) kbps[30 40) kbps [40 50) kbps and [50 60] kbps Furthermoreif the throughput value of ws119895 observed by user119894 is 44 kbps inWS-DREAM then the transformed user-service rating thatis 119877119894minus119895 = 4lowast holds For each service invoked by a user werandomly select a time interval from all the 64 intervals and



(d)

(c)(b)

Preferredservices

Service recommendation

(a)

Usertarget Useri

Userg

Usery

Userx

Userk

Userz

middot middot middot



Enemy_set (usertarget)

Friend_set (useri)

Enemy_set (useri)

Indirect_friend_set (usertarget)

Figure 3 Relationships among usertarget user119894 user119896 user119892 user119909 user119910 and user119911

transform the service quality corresponding to the selectedtime interval into a user-service rating with a timestamp (iethe selected time interval) In the experiment only a qualitydimension that is throughput is considered

Besides our paper only discusses the time-aware servicerecommendation in sparse-data environment where targetuser has no similar friends and similar services Thereforewe select appropriate data fromWS-DREAM to simulate theabove sparse recommendation situations Furthermore foreach recruited target user hisher 80 ratings are randomlyselected for training while the remaining 20 ratings fortesting

Also we compare our proposal with related works forexample WSRec [25] MCCP [26] and SBT-SR [27] interms of recommendation accuracy and recall Concretelyaccuracy and recall are measured by Mean Absolute Error(ie MAE the smaller the better) in (13) and Recall (thelarger the better) in (14) In (13) Rec Ser Set denotes therecommended service set for target user |119883| denotes the sizeof set 119883 119877targetminus119909 and 119877targetminus119909 represent target userrsquos realand predicted ratings over service ws119909 respectively While in(14) symbols Preferred Ser Set and Rec Ser Set denote theservice set really preferred (with 4lowast or 5lowast rating) by target userand the service set recommended to target user respectively

MAE = sumws119909isinRec Ser Set

10038161003816100381610038161003816119877targetminus119909 minus 119877targetminus11990910038161003816100381610038161003816|Rec Ser Set| (13)

Recall = |Preferred Ser Set cap Rec Ser Set||Preferred Ser Set| (14)

The experiments were conducted on a Dell laptop with280GHz processors and 20GB RAM The software config-uration is Windows XP + JAVA 15 Each experiment wascarried out 10 times and the average results were adoptedfinally

42 Experiment Results and Analyses Concretely four pro-files are designed tested and comparedHere119898 and 119899denotethe number of users and number of services in the historicaluser-service invocation network

(Profile 1) Recommendation Accuracy Comparison In thisprofile we compare the recommendation accuracy ofSer Rectime with the other three approaches that is WSRecMCCP and SBT-SR Concretely parameters 119908119906 = 119908119894 = 05hold in WSRec 120572 = 08 holds in MCCP user similaritythreshold 119875 = minus05 holds in both SBT-SR and Ser RectimeBesides in Ser Rectime 120572 = 009 holds in (3) and 119882load =119882version = 1 holds (in WS-DREAM a service was invokedby a user 64 times within successive 16 hours therefore weassume that the service load and version do not vary much)Finally119898 is varied from 20 to 100 in all the four approaches

The experiment result is presented in Figure 4 AsFigure 4 shows the recommendation accuracy of WSRecis low as it only adopts the average ratings of target userand target services without considering the valuable user-service relationships hidden in historical service invocationrecords The accuracy of MCCP is improved by consideringthe user-service relationships however it is not very suitablefor service recommendation when target user has no similarfriends and similar services as MCCP is essentially a kind


Inputs(1) 119880 = user1 user119898 a set of users in user-service invocation amp rating network(2) WS = ws1 ws119899 a set of web services in user-service invocation amp rating network


ws119895) | 1 le 119894 le 119898 1 le 119895 le 119899 a set of historical user-service rating records(4) usertarget target user who requires service recommendationOutput Rec Ser Set service set recommended to usertarget(1) Set user similarity threshold 119875 (minus1 le 119875 le minus05)(2) for each user119894 isin 119880 do Step 1(3) calculate Simtime(usertarget user119894) based on (1)ndash(6)(4) if Simtime(usertarget user119894) le 119875(5) then put user119894 into set Enemy set(usertarget)(6) end if(7) end for(8) for each user119894 isin Enemy set(usertarget) do Step 2(9) determine user119894rsquos friend user119892 based on (6) and (8)(10) if probability in (10) is larger than minus119875(11) then put user119892 into Enemy set(usertarget)(12) end if(13) determine user119894rsquos enemy user119896 based on (6) and (7)(14) if probability in (9) is larger than minus119875(15) then put user119896 into Indirect friend set(usertarget)(16) end if(17) end for(18) for each user119909 isin Indirect friend set(usertarget) do(19) determine user119909rsquos enemy user119910 based on (6) and (7)(20) if probability in (11) is larger than minus119875(21) then put user119910 into Enemy set(usertarget)(22) end if(23) determine user119909rsquos friend user119911 based on (6) and (8)(24) if probability in (12) is larger than minus119875(25) then put user119911 into Indirect friend set(usertarget)(26) end if(27) end for(28) repeat Line (8)ndashLine (27) until Indirect friend set(usertarget) stays stable(29) Rec Ser Set = 0 Step 3(30) for each user119909 isin Indirect friend set(usertarget) do(31) for each ws119895 isinWS do(32) if 119877119909minus119895 isin 4lowast 5lowast(33) then put ws119895 into set Rec Ser Set(34) end if(35) end for(36) end for(37) return Rec Ser Set

Algorithm 1 Ser Rectime (119880WSrarr usertarget)

of similar user-based service recommendation approachTherecommendation accuracy of SBT-SR is high as ldquoenemyrsquosenemy is a friendrdquo rule of Social Balance Theory is recruitedto find the ldquoindirect friendsrdquo of target user Furthermoreour proposed Ser Rectime outperforms SBT-SR as Ser Rectimenot only considers Social Balance Theory but also takestime factor into consideration for looking for the reallysimilar ldquoindirect friendsrdquo of target user Besides as shownin Figure 4 recommendation accuracy values of MCCPSBT-SR and Ser Rectime all increase with the growth of 119898approximately this is because more valuable user-servicerelationship information would be discovered and recruited

for service recommendation when there are many users aswell as their invocation records

(Profile 2) Recommendation Recall Comparison In this pro-file we compare the recommendation recall values of fourapproaches The parameter settings are the same as those inprofile 1 Concrete experiment results are shown in Figure 5

As Figure 5 shows the recall of WSRec is low as theldquoaveragerdquo idea adopted in WSRec often leads to a low recom-mendation hit rate The recommendation recall values of theremaining three approaches all increase with the growth of119898this is because when there are many available historical users


1

11

12

13

14

15

16

17

18

19

2

MA

E

40 60 80 10020m number of users

WSRecSer_Rectime SBT-SR

MCCP

Figure 4 Recommendation accuracy comparison with respect to119898

5

10

15

20

25

30

35

40

45

50

Reca

ll (

)



MCCP

Figure 5 Recommendation recall comparison with respect to119898

more useful user-service relationships could be mined andrecruited in service recommendation hence more qualifiedrecommendation results are returned finally However therecall of MCCP is still not high as few really similar friends oftarget user could be found in the recommendation situationsthat we discuss in this paper (ie the sparse recommendationsituations when target user has no similar friends and similarservices)

While both SBT-SR and Ser Rectime approaches achievegood performances in recommendation recall as SocialBalance Theory is recruited to find out the indirect friendsof target user even if the target user has no similar friends

0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)



MCCP

Figure 6 Time cost comparison with respect to119898

and similar services Furthermore our proposed Ser Rectimeapproach often outperforms SBT-SR in recall This is becauseonly ldquoenemyrsquos enemy is a friendrdquo rule of Social BalanceTheory is recruited in SBT-SR while in Ser Rectime allfour rules in Figure 2 are considered which improves therecommendation hit rate to some extent

(Profile 3) Execution Efficiency Comparison with respect to119898 In this profile we test the execution efficiency of fourapproaches with respect to the number of users ie119898 Here119898 is varied from 20 to 100 the number of services that is119899 = 1000 holds besides 119875 = minus05 120572 = 009 and 119882load =119882version = 1 holdThe experiment result is shown in Figure 6

As Figure 6 shows time cost of WSRec is the best asWSRec only adopts the average ratings of target user andtarget services without complex computationThe time costsof the remaining three approaches all increase with thegrowth of 119898 quickly as more similarity computation costis required to determine the similar friends or dissimilarenemies of a user when the number of users increasesFurthermore Ser Rectime often requires more computationtime as multiple iteration processes are probable for findingall the indirect friends of target user However as Figure 6indicates the execution efficiency of Ser Rectime is oftenacceptable (at ldquominuterdquo level)

(Profile 4) Execution Efficiency Comparison with respect to119899 In this profile we test the execution efficiency of fourapproaches with respect to the number of services that is119899 Here 119899 is varied from 200 to 1000 the number of usersthat is 119898 = 100 holds besides 119875 = minus05 120572 = 009 and119882load = 119882version = 1 hold The concrete experiment result isshown in Figure 7

As Figure 7 shows similar to profile 3 the time cost ofWSRec is the best due to the adopted average idea Besides


200 400 600 800 1000n number of services

0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)


MCCP

Figure 7 Time cost comparison with respect to 119899

the time costs of the remaining three approaches all increasewith the growth of 119898 approximately linearly as each serviceis considered at most once in each user similarity calculationprocess However as can be seen from Figure 7 the servicerecommendation process of Ser Rectime approach could begenerally finished in polynomial time

5 Evaluations

In this section we first analyze the time complexity of ourproposed Ser Rectime approach Afterwards related worksand comparison analyses are presented to further clarify theadvantages and application scope of our proposal Finally wepoint out our future research directions

51 Complexity Analyses Suppose there are 119898 users and 119899services in the historical user-service invocation network

Step 1 According to (1)ndash(6) the time-aware similaritybetween target user and any other user could be calculatedwhose time complexity is 119874(119899) Afterwards a user is judgedto be an qualified enemy of target user or not based on (7)whose time complexity is 119874(1) As there are totally 119898 usersin set 119880 the time complexity of this step is 119874(119898 lowast (119899 + 1)) =119874(119898 lowast 119899)Step 2 In this step we utilize the four rules (see Figure 3) inSocial BalanceTheory iteratively so as to find all the indirectfriends of target user And in the worst case the similaritybetween any two users in set 119880 needs to be calculated Asthere are 119898 users in set 119880 the time complexity of this stepis 119874(1198982 lowast 119899)Step 3 For each derived indirect friend (atmost119898minus1 indirectfriends) of target user we select hisher preferred services (at

most 119899 services) and recommend them to the target user Asthe time complexity of preferred-service-judgment process is119874(1) the time complexity of this step is 119874(119898 lowast 119899)

With the above analyses we can conclude that the totaltime complexity of our proposed Ser Rectime approach is119874(1198982 lowast 119899)52 Related Works and Comparison Analyses With theadvent of IoE age an excessive number of IoE services areemerging on the web which places a heavy burden on theservice selection decision of target users In this situationvarious recommendation techniques for example CF-basedrecommendation [28] and content-based recommendation[29] are put forward to help the target users find theirinterested IoE services

A two-phase 119870-means clustering approach is broughtforth in [30] to make service quality prediction and servicerecommendation However this clustering-based approachoften requires a dense historical user-service invocationmatrix and hence cannot deal with the service recommen-dation problem on sparse data very well In [31] a CF-basedrecommendation approach is put forward which realizes ser-vice recommendation based on the similar friends of targetuser However when the target user has no similar friendsthe recommendation accuracy is decreased significantly Abidirectional (ie user-based CF + item-based CF) servicerecommendation approach named WSRec is brought forthin [25] for high-quality recommendation results Howeverwhen a target user has no similar friends and similarservices WSRec can only make a rough prediction andrecommendation by considering both the average ratingfrom target user and the average rating of the service thatis ready to be predicted In [26] a MCCP approach is putforward to capture and model the preferences of varioususers over different services however only similar friends oftarget user are recruited for service quality prediction andrecommendation which drops some valuable user-servicerelationship information hidden in the historical user-serviceinvocation records

In order to mine and introduce more user-service rela-tionship information into recommendation a service rec-ommendation approach SBT-SR is proposed in our previouswork [27] By utilizing ldquoenemyrsquos enemy is a friendrdquo rule inSocial Balance Theory some indirect friends of target usercould be found and utilized for further recommendationwhich improves the accuracy and recall of recommendationin sparse-data environment However SBT-SR has two short-comings first service invocation time is not considered inSBT-SR which may decrease the recommendation accuracyas service quality is often dynamic and varied with timesecond SBT-SR only employs ldquoenemyrsquos enemy is a friendrdquorule for service recommendation while overlooks othervaluable rules in Social BalanceTheory for example ldquofriendrsquosfriend is a friendrdquo rule ldquofriendrsquos enemy is an enemyrdquo rule andldquoenemyrsquos friend is an enemyrdquo rule In view of the above twoshortcomings a novel time-aware service recommendationapproach named Ser Rectime is put forward in this paperto deal with the recommendation problem in sparse-data


environment Ser Rectime considers not only the serviceinvocation time but also the four rules of Social BalanceThe-ory so that the recommendation accuracy and recall could beensured Finally through a set of experiments deployed ona real web service quality dataset WS-DREAM we validatethe feasibility of Ser Rectime in terms of recommendationaccuracy recall and efficiency

53 FurtherDiscussions In this paper we put forward a time-aware similarity for service recommendation Generally theproposed time-aware similarity could also be applied inother similarity-based application domains such as contentsearching [32ndash36] information detection [37ndash45] and qual-ity optimization [46ndash50] However there are still severalshortcomings in our paper which are discussed as follows

(1) The time cost of our proposed Ser Rectime approachincreases fast when the number of users growsThere-fore the execution efficiency of Ser Rectime needs tobe improved especially when a huge number of usersare present in the historical user-service invocationrecords

(2) In this paper we have investigated the time-awareuser similarity However besides service invocationtime many other factors for example user-servicelocation information also play an important rolein user similarity calculation In the future we willimprove our proposal by combining the time andlocation factors together

6 Conclusions

In this paper a novel time-aware service recommendationapproach that is Ser Rectime is put forward to handlethe service recommendation problems in sparse-data envi-ronment where target user has no similar friends andsimilar services Instead of looking for similar friends intraditional CF-based service recommendation approaches inSer Rectime we first look for dissimilar enemies of target userbased on time-aware user similarity and further determinethe indirect friends of target user based on Social BalanceTheory Afterwards the services preferred by indirect friendsof target user are recommended to the target user Finallythrough a set of experiments deployed on a real web servicequality datasetWS-DREAM we validate the feasibility of ourproposal in terms of recommendation accuracy recall andefficiency

In the future we will improve the recommendation effectof our proposal by considering more user-service locationinformation Moreover distributed or parallel recommenda-tion approaches will be investigated in the future to improvethe recommendation efficiency

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This paper is partially supported by Natural Science Foun-dation of China (no 61402258 no 61602253 no 61672276no 61373027 and no 61672321) and Open Project ofState Key Laboratory for Novel Software Technology (noKFKT2016B22)

References

[1] Y Duan G Fu N Zhou X Sun N C Narendra and B HuldquoEverything as a service (XaaS) on the cloud origins currentand future trendsrdquo in Proceedings of the 8th InternationalConference on Cloud Computing (Cloud rsquo15) pp 621ndash628 NewYork NY USA July 2015

[2] Y Ren J Shen J Wang J Han and S Lee ldquoMutual verifiableprovable data auditing in public cloud storagerdquo Journal ofInternet Technology vol 16 no 2 pp 317ndash323 2015

[3] J Shen H Tan J Wang J Wang and S Lee ldquoA novel routingprotocol providing good transmission reliability in underwatersensor networksrdquo Journal of Internet Technology vol 16 no 1pp 171ndash178 2015

[4] C Yuan X Sun and L V Rui ldquoFingerprint liveness detectionbased on multi-scale LPQ and PCArdquo China Communicationsvol 13 no 7 pp 60ndash65 2016

[5] X Wen L Shao Y Xue and W Fang ldquoA rapid learningalgorithm for vehicle classificationrdquo Information Sciences vol295 no 1 pp 395ndash406 2015

[6] Z Xia X Wang X Sun and Q Wang ldquoA secure and dynamicmulti-keyword ranked search scheme over encrypted clouddatardquo IEEETransactions onParallel andDistributed Systems vol27 no 2 pp 340ndash352 2016

[7] Z Fu X Sun Q Liu L Zhou and J Shu ldquoAchieving effi-cient cloud search services multi-keyword ranked search overencrypted cloud data supporting parallel computingrdquo IEICETransactions on Communications vol 98 no 1 pp 190ndash2002015

[8] Z Xia X Wang L Zhang Z Qin X Sun and K Ren ldquoAprivacy-preserving and copy-deterrence content-based imageretrieval scheme in cloud computingrdquo IEEE Transactions onInformation Forensics and Security vol 11 no 11 pp 2594ndash26082016

[9] Z Pan Y Zhang and S Kwong ldquoEfficient motion and disparityestimation optimization for low complexity multiview videocodingrdquo IEEE Transactions on Broadcasting vol 61 no 2 pp166ndash176 2015

[10] S Xie and Y Wang ldquoConstruction of tree network with limiteddelivery latency in homogeneous wireless sensor networksrdquoWireless Personal Communications vol 78 no 1 pp 231ndash2462014

[11] Z Zhou Y Wang J Wu C N Yang and X Sun ldquoEffective andefficient global context verification for image copy detectionrdquoIEEE Transactions on Information Forensics and Security vol 12no 1 pp 48ndash63 2016

[12] L Qi X Xu X Zhang et al ldquoStructural Balance Theory-based E-commerce recommendation over big rating datardquo IEEETransactions on Big Data 2016

[13] T Ma J Zhou M Tang et al ldquoSocial network and tag sourcesbased augmenting collaborative recommender systemrdquo IEICETransactions on Information and Systems vol 98 no 4 pp 902ndash910 2015


[14] L Qi W Dou C Hu Y Zhou and J Yu ldquoA context-aware ser-vice evaluation approach over big data for cloud applicationsrdquoIEEE Transactions on Cloud Computing

[15] X Fan Y Hu R Zhang W Chen and P Brezillon ldquoModelingtemporal effectiveness for context-aware web services recom-mendationrdquo in Proceedings of the IEEE International Conferenceon Web Services (ICWS rsquo15) pp 225ndash232 New York NY USAJune 2015

[16] S Wang L Huang C-H Hsu and F Yang ldquoCollaborationreputation for trustworthy Web service selection in socialnetworksrdquo Journal of Computer and System Sciences vol 82 no1 pp 130ndash143 2016

[17] S Wang Y Ma B Cheng F Yang and R Chang ldquoMulti-dimensional QoS prediction for service recommendationsrdquoIEEE Transaction on Services Computing 2016

[18] Y Ma S Wang P C Hung C H Hsu Q Sun and F Yang ldquoAhighly accurate prediction algorithm for unknown web serviceQoS valuerdquo IEEE Transactions on Services Computing vol 9 no4 pp 511ndash523 2016

[19] S Wang L Huang L Sun C-H Hsu and F Yang ldquoEfficientand reliable service selection for heterogeneous distributedsoftware systemsrdquo Future Generation Computer Systems 2016

[20] Y Ni Y Fan W Tan K Huang and J Bi ldquoNCSR negative-connection-aware service recommendation for large sparseservice networkrdquo IEEE Transactions on Automation Science andEngineering vol 13 no 2 pp 579ndash590 2016

[21] Y Zhong Y Fan K Huang W Tan and J Zhang ldquoTime-aware service recommendation for mashup creationrdquo IEEETransactions on Services Computing vol 8 no 3 pp 356ndash3682015

[22] D Cartwright and F Harary ldquoStructural balance a generaliza-tion of Heiderrsquos theoryrdquo Psychological Review vol 63 no 5 pp277ndash293 1956

[23] Z Zheng Y Zhang and M R Lyu ldquoInvestigating QoS of real-world web servicesrdquo IEEE Transactions on Services Computingvol 7 no 1 pp 32ndash39 2014

[24] B Sarwar G Karypis J Konstan and J Reidl ldquoItem-based col-laborative filtering recommendation algorithmsrdquo inProceedingsof the the 10th International Conference onWorld WideWeb pp285ndash295 ACM Hong Kong May 2001

[25] Z Zheng H Ma M R Lyu and I King ldquoQoS-aware webservice recommendation by collaborative filteringrdquo IEEETrans-actions on Services Computing vol 4 no 2 pp 140ndash152 2011

[26] Y Rong X Wen and H Cheng ldquoA Monte Carlo algorithmfor cold start recommendationrdquo in Proceedings of the 23rdInternational Conference on World Wide Web (WWW rsquo14) pp327ndash336 ACM Seoul South Korea April 2014

[27] L Qi X Zhang Y Wen and Y Zhou ldquoA Social BalanceTheory-based service recommendation approachrdquo in Advancesin Services Computing 9th Asia-Pacific Services ComputingConference APSCC 2015 Bangkok Thailand December 7ndash92015 Proceedings vol 9464 of Lecture Notes in ComputerScience pp 48ndash60 Springer Berlin Germany 2015

[28] F Zhang T Gong V E Lee G Zhao C Rong and G Qu ldquoFastalgorithms to evaluate collaborative filtering recommendersystemsrdquo Knowledge-Based Systems vol 96 pp 96ndash103 2016

[29] L Yao Q Z Sheng A H H Ngu J Yu and A Segev ldquoUnifiedcollaborative and content-based web service recommendationrdquoIEEE Transactions on Services Computing vol 8 no 3 pp 453ndash466 2015

[30] C Wu W Qiu Z Zheng X Wang and X Yang ldquoQoS predic-tion of web services based on two-phase K-means clusteringrdquoin Proceedings of the IEEE International Conference on WebServices (ICWS rsquo15) pp 161ndash168 IEEE New York NY USA July2015

[31] S-Y Lin C-H Lai C-H Wu and C-C Lo ldquoA trustworthyQoS-based collaborative filtering approach for web servicediscoveryrdquo Journal of Systems and Software vol 93 pp 217ndash2282014

[32] Z Fu X Wu C Guan X Sun and K Ren ldquoTowards efficientmulti-keyword fuzzy search over encrypted outsourced datawith accuracy improvementrdquo IEEE Transactions on InformationForensics and Security vol 11 no 12 pp 2706ndash2716 2016

[33] Z Fu K Ren J Shu X Sun and F Huang ldquoEnabling person-alized search over encrypted outsourced data with efficiencyimprovementrdquo IEEE Transactions on Parallel and DistributedSystems vol 27 no 9 pp 2546ndash2559 2016

[34] Z Pan P Jin J Lei Y Zhang X Sun and S Kwong ldquoFastreference frame selection based on content similarity for lowcomplexity HEVC encoderrdquo Journal of Visual Communicationand Image Representation vol 40 pp 516ndash524 2016

[35] Z Fu F Huang X Sun A V Vasilakos and C Yang ldquoEnablingsemantic search based on conceptual graphs over encryptedoutsourced datardquo IEEE Transactions on Services Computing2016

[36] Z Fu X Sun S Ji and G Xie ldquoTowards efficient content-awaresearch over encrypted outsourced data in cloudrdquo in Proceedingsof the 35th Annual IEEE International Conference on ComputerCommunications (INFOCOM rsquo16) pp 1ndash9 San Francisco CalifUSA April 2016

[37] Z Xia X Wang X Sun and B Wang ldquoSteganalysis of leastsignificant bit matching using multi-order differencesrdquo Securityand Communication Networks vol 7 no 8 pp 1283ndash1291 2014

[38] J Li X Li B Yang and X Sun ldquoSegmentation-based imagecopy-move forgery detection schemerdquo IEEE Transactions onInformation Forensics and Security vol 10 no 3 pp 507ndash5182015

[39] Z Pan J Lei Y Zhang X Sun and S Kwong ldquoFast motionestimation based on content property for low-complexityH265HEVC encoderrdquo IEEE Transactions on Broadcasting vol62 no 3 pp 675ndash684 2016

[40] Y Zheng B Jeon D Xu Q M J Wu and H Zhang ldquoImagesegmentation by generalized hierarchical fuzzy C-means algo-rithmrdquo Journal of Intelligent and Fuzzy Systems vol 28 no 2pp 961ndash973 2015

[41] B Chen H Shu G Coatrieux G Chen X Sun and J L Coa-trieux ldquoColor image analysis by quaternion-type momentsrdquoJournal of Mathematical Imaging and Vision vol 51 no 1 pp124ndash144 2015

[42] Z Xia X Wang X Sun Q Liu and N Xiong ldquoSteganalysisof LSBmatching using differences between nonadjacent pixelsrdquoMultimedia Tools and Applications vol 75 no 4 pp 1947ndash19622016

[43] Y Zhang X Sun and B Wang ldquoEfficient algorithm for k-barrier coverage based on integer linear programmingrdquo ChinaCommunications vol 13 no 7 pp 16ndash23 2016

[44] JWang T Li Y Shi S Lian and J Ye ldquoForensics feature analysisin quaternion wavelet domain for distinguishing photographicimages and computer graphicsrdquoMultimedia Tools and Applica-tions 2016

[45] Z Zhou C Yang B Chen X Sun Q Liu and Q J WuldquoEffective and efficient image copy detection with resistance


to arbitrary rotationrdquo IEICE Transactions on Information andSystems D vol 99 no 6 pp 1531ndash1540 2016

[46] L Qi W Dou and J Chen ldquoWeighted principal componentanalysis-based service selection method for multimedia ser-vices in cloudrdquo Computing vol 98 no 1-2 pp 195ndash214 2016

[47] Y Chen C Hao W Wu and E Wu ldquoRobust dense reconstruc-tion by range merging based on confidence estimationrdquo ScienceChina Information Sciences vol 59 no 9 Article ID 092103 11pages 2016

[48] Q LiuW Cai J Shen Z Fu X Liu andN Linge ldquoA speculativeapproach to spatialminustemporal efficiency with multiminusobjectiveoptimization in a heterogeneous cloud environmentrdquo Securityand Communication Networks vol 9 no 17 pp 4002ndash40122016

[49] Y Kong M Zhang and D Ye ldquoA belief propagation-basedmethod for task allocation in open and dynamic cloud environ-mentsrdquo Knowledge-Based Systems vol 115 pp 123ndash132 2017

[50] B Gu V S Sheng and S Li ldquoBi-parameter space partitionfor cost-sensitive SVMrdquo in Proceedings of the 24th InternationalConference on Artificial Intelligence (ICAI rsquo15) pp 3532ndash3539Las Vegas Nev USA July 2015

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



user-service rating data is sparse Concretely in Ser Rectimewe first calculate the time-aware user similarity afterwardswe look for the enemies (antonym of ldquofriendrdquo) of the targetuser and further determine the ldquoindirect friendsrdquo of targetuser by Social Balance Theory [22] (eg ldquoenemyrsquos enemy is afriendrdquo rule) finally the services preferred by indirect friendsof target user are recommended to the target user

The contributions of our paper are threefold

(1) Time factor is considered in user similarity calcula-tion to adapt to the dynamic quality variation of IoEservices which makes the subsequent recommenda-tion result more objective and accurate

(2) Social Balance Theory is introduced for service rec-ommendation on sparse data so that the recommen-dation robustness is improved

(3) A wide range of experiments are designed anddeployed on a real web service quality set WS-DREAM [23] so as to further validate the feasibilityof our proposal

The remainder of our paper is organized as follows InSection 2 we first formalize the CF-based service recommen-dation problem and afterwards demonstrate the motivationof our paper In Section 3 a novel time-aware servicerecommendation approach named Ser Rectime is broughtforth A set of experiments are deployed in Section 4 andevaluations are presented in Section 5 Finally in Section 6we summarize the paper and point out our future researchdirections

2 Formalization and Motivation

In this section we first formalize theCF-based service recom-mendation problemAnd afterwards an example is presentedto demonstrate the motivation of our paper intuitively

21 Formal Specification Generally the CF-based servicerecommendation problem could be specified with a four-tuple CF-Ser-Rec (119880WSrarr usertarget) where

(1) 119880 = user1 user119898 denotes the user set in user-service invocation network and 119898 is the number ofusers

(2) WS = ws1 ws119899 denotes the web service set inuser-service invocation network and 119899 is the numberof web services


ws119895) | 1 le 119894 le 119898 1 le 119895 le 119899 de-notes the historical invocation record set that isuser119894

119877119894minus119895997888997888997888rarr119879119894minus119895

ws119895 whichmeans that user119894 invoked ws119895 in

the past and the rating of ws119895 by user119894 is119877119894minus119895 Here forsimplicity the well-known 1lowast 2lowast 3lowast 4lowast 5lowast ratingsystem is adopted to depict119877119894minus119895 Also each invocationrecord owns a timestamp 119879119894minus119895 that indicates the ser-vice invocation time (here we assume the timestampformat is ldquoyyyymmddrdquo)

(4) usertarget denotes the target user that requires servicerecommendation and usertarget isin 119880 holds

With the above formalization CF-based service recom-mendation problem could be specified as follows accordingto the historical invocation record set ldquorarrrdquo between users (in119880) and web services (in WS) find out the services that werenever invoked but may be preferred by target user usertargetand recommend them to usertarget

22 Motivation In this subsection the paper motivation isclarified more intuitively with the example shown in Figure 1In Figure 1 there are three users TomAliceBob (Tom isthe target user) in set 119880 and six web services ws1ws2ws3ws4ws5ws6 in set WS historical user-service invocationrecord set rarr (including rating data and timestamp) is alsopresented in Figure 1

Then according to the Adjusted Cosine Similarity [24](ie ACS isin [minus1 1] here the reason that we utilize ACSfor similarity calculation is that user-service rating data isoften discrete and the rating scales of different users oftenvary) we can calculate the similarity between target userTom and other two users (ie Alice and Bob) ConcretelySim(TomAlice) = minus02747 and Sim(TomBob) = Null (asno services were rated by both Tom and Bob) Therefore wecan conclude that target user Tom has no similar friends soaccording to the traditional user-based CF recommendationapproaches no qualified services are recommended to Tom

Likewise the similarities between target services (ie ws1andws2) and other services (ie ws3 ws4 ws5 andws6) couldalso be obtained Concretely Sim(ws1ws3) = Sim(ws1ws4)= Sim(ws2ws3) = Sim(ws2ws4) = minus1 and Sim(ws1ws5) =Sim(ws1ws6) = Sim(ws2ws5) = Sim(ws2ws6) = Null (asno users invoked any pair of services above simultaneously)Therefore a conclusion could be drawn that target user Tomrsquospreferred services (ie ws1 and ws2) have no similar servicesso according to the traditional item-based CF recommenda-tion approaches no qualified services are recommended toTom

Therefore in this situation traditional CF-based ser-vice recommendation approaches (eg user-based CF item-based CF or hybrid CF) cannot make accurate service rec-ommendation for target user Besides as Figure 1 shows thetimestamps of various invocation records are often differentso the service recommendation accuracy and fairness wouldbe decreased if we overlook the time factor in recommen-dation process In view of the above two challenges a novelservice recommendation approach named Ser Rectime is putforward in Section 3 Ser Rectime cannot only find outthe indirect friends of a target user so as to improve therecommendation robustness in sparse-data environment butalso consider time factor in recommendation so as to ensurethe fairness and accuracy of recommendation results

3 Time-Aware Service RecommendationApproach Ser_Rectime

In this section a novel time-aware recommendation ap-proach that is Ser Rectime is introduced to deal with the


Tom Alice Bob

User level

Service level

20100101 20120407 20110217

20140927 20130722

20151126 20150116

20130616 2016011420160609

Usertarget

1lowast

1lowast

1lowast1lowast

2lowast

4lowast5lowast

5lowast

5lowast

5lowast











(1)















119882version


0 else(5)



(6)















(9)


P O


X

(a)

P O


X

(b)

X

P O


(c)

X

P O


(d)



(10)



(11)


(12)




4 Experiment






(d)

(c)(b)

Preferredservices


(a)

Usertarget Useri

Userg

Usery

Userx

Userk

Userz





Friend_set (useri)

Enemy_set (useri)























1

11

12

13

14

15

16

17

18

19

2

MA

E



MCCP


5

10

15

20

25

30

35

40

45

50

Reca

ll (

)



MCCP




0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)



MCCP









0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)


MCCP



5 Evaluations













6 Conclusions



Competing Interests


Acknowledgments


References





























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Tom Alice Bob

User level

Service level

20100101 20120407 20110217

20140927 20130722

20151126 20150116

20130616 2016011420160609

Usertarget

1lowast

1lowast

1lowast1lowast

2lowast

4lowast5lowast

5lowast

5lowast

5lowast











(1)















119882version


0 else(5)



(6)















(9)


P O


X

(a)

P O


X

(b)

X

P O


(c)

X

P O


(d)



(10)



(11)


(12)




4 Experiment






(d)

(c)(b)

Preferredservices


(a)

Usertarget Useri

Userg

Usery

Userx

Userk

Userz





Friend_set (useri)

Enemy_set (useri)























1

11

12

13

14

15

16

17

18

19

2

MA

E



MCCP


5

10

15

20

25

30

35

40

45

50

Reca

ll (

)



MCCP




0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)



MCCP









0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)


MCCP



5 Evaluations













6 Conclusions



Competing Interests


Acknowledgments


References





























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in









119882version


0 else(5)



(6)















(9)


P O


X

(a)

P O


X

(b)

X

P O


(c)

X

P O


(d)



(10)



(11)


(12)




4 Experiment






(d)

(c)(b)

Preferredservices


(a)

Usertarget Useri

Userg

Usery

Userx

Userk

Userz





Friend_set (useri)

Enemy_set (useri)























1

11

12

13

14

15

16

17

18

19

2

MA

E



MCCP


5

10

15

20

25

30

35

40

45

50

Reca

ll (

)



MCCP




0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)



MCCP









0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)


MCCP



5 Evaluations













6 Conclusions



Competing Interests


Acknowledgments


References





























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




P O


X

(a)

P O


X

(b)

X

P O


(c)

X

P O


(d)



(10)



(11)


(12)




4 Experiment






(d)

(c)(b)

Preferredservices


(a)

Usertarget Useri

Userg

Usery

Userx

Userk

Userz





Friend_set (useri)

Enemy_set (useri)























1

11

12

13

14

15

16

17

18

19

2

MA

E



MCCP


5

10

15

20

25

30

35

40

45

50

Reca

ll (

)



MCCP




0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)



MCCP









0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)


MCCP



5 Evaluations













6 Conclusions



Competing Interests


Acknowledgments


References





























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





(d)

(c)(b)

Preferredservices


(a)

Usertarget Useri

Userg

Usery

Userx

Userk

Userz





Friend_set (useri)

Enemy_set (useri)























1

11

12

13

14

15

16

17

18

19

2

MA

E



MCCP


5

10

15

20

25

30

35

40

45

50

Reca

ll (

)



MCCP




0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)



MCCP









0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)


MCCP



5 Evaluations













6 Conclusions



Competing Interests


Acknowledgments


References





























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in













1

11

12

13

14

15

16

17

18

19

2

MA

E



MCCP


5

10

15

20

25

30

35

40

45

50

Reca

ll (

)



MCCP




0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)



MCCP









0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)


MCCP



5 Evaluations













6 Conclusions



Competing Interests


Acknowledgments


References





























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




1

11

12

13

14

15

16

17

18

19

2

MA

E



MCCP


5

10

15

20

25

30

35

40

45

50

Reca

ll (

)



MCCP




0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)



MCCP









0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)


MCCP



5 Evaluations













6 Conclusions



Competing Interests


Acknowledgments


References





























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





0

02

04

06

08

1

12

14

16

Tim

e cos

t (m

in)


MCCP



5 Evaluations













6 Conclusions



Competing Interests


Acknowledgments


References





























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in








6 Conclusions



Competing Interests


Acknowledgments


References





























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in


















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in

















Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in










Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



Documents

Research Article Time-Aware IoE Service Recommendation on …downloads.hindawi.com/journals/misy/2016/4397061.pdf · 2019-07-30 · 2016.01.14 2016.06.09 User target 1 1 1 1 2 5 4